Toolkit
What I've used, and where.
A list of skills is easy to pad. So this one only has tools that are in a public repository of mine, marked against the project that uses them. Click a column to see the project.
RD Reasonable DoubtHM Honest Mistake
PE Predictive EngineRZ RegretZero
| Tool | RD | HM | PE | RZ |
|---|---|---|---|---|
| Modelling | ||||
| XGBoost | · | ● | · | · |
| LightGBM | · | · | ● | ● |
| Quantile regression | · | · | · | ● |
| SHAP | · | ● | · | · |
| Optuna | · | ● | · | · |
| scikit-learn | ● | ● | · | ● |
| LLMs & agents | ||||
| Fine-tuning a transformer (DeBERTa-v3) | ● | · | · | · |
| QLoRA | ● | · | · | · |
| Anthropic API | ● | ● | · | · |
| ReAct tool-use loop | · | ● | · | · |
| Embeddings + pgvector retrieval | · | ● | · | · |
| Prompt caching | · | ● | · | · |
| Serving & infra | ||||
| ONNX Runtime | ● | · | · | · |
| FastAPI | ● | · | · | · |
| Docker | ● | ● | · | · |
| Google Cloud Run | ● | · | · | · |
| Streamlit | · | · | · | ● |
| PostgreSQL | · | ● | · | · |
| GitHub Actions / pytest | ● | · | · | · |
| Quant & decisions | ||||
| Walk-forward backtesting | · | · | ● | · |
| NautilusTrader | · | · | ● | · |
| Transaction cost & tax modelling | · | · | ● | · |
| Newsvendor optimisation | · | · | · | ● |
| Evaluation discipline | ||||
| Temporal holdouts | · | ● | ● | ● |
| Leakage audits | · | ● | ● | ● |
| Bootstrap confidence intervals | ● | · | · | · |
| Pre-registered experiments | ● | ● | ● | · |
Underneath
Languages
Python for everything above. SQL on PostgreSQL. Enough JavaScript to build Honest Mistake's in-browser audit page without a framework.
Mathematics
My degree is in Data and Computational Science with a mathematics specialisation: probability, statistics, linear algebra and optimisation are the coursework, not a side interest.
Deep learning
PyTorch, backpropagation, attention and transformers, learned from the maths up and put to use when fine-tuning DeBERTa for Reasonable Doubt.